12 Replies Latest reply on Nov 4, 2010 2:10 PM by Harm Millaard

    Odds and ends

    ChrissaSwenson

      Just for everyone's info:

       

      Read about the guy that had a "patched" GPU system (uncertified card edited into text file "fix") and it wouldn't Direct Export properly without crashing PPro.  My friend's PC did the same thing with WMV and MOV exports with a GTX-9600 card.  Wonder if that can be explained, technically.  Happened with both old and new NVIDIA drivers loaded.

       

      By the way, they latest NVIDIA drivers brag of GPU support improvements for Premiere Pro.  What are those?  It acts the same.

       

      Notice that my playback in PPro from timeline is ragged and sliced at times with the latest PPro upgrade.  Before then it looked beautiful.  So, better GPU support at the cost of poorer display quality in timeline viewing?

       

      AME still encodes 3-5X slower than Direct Export.  Have any of you been told this was fixed or an old issue recently?

       

      I'm still using this software, but I'm not impressed with certain things about it and the lack of answer from the top.

       

       

      Best to all,

       

      Carissa

        • 1. Re: Odds and ends
          Ann Bens Adobe Community Professional & MVP

          Maybe you should use a certified card instead of a

          hacked one.

          • 2. Re: Odds and ends
            Harm Millaard Level 7

            Let's clear up a few things first:

             

            Is it CHrissa or CArissa? You have me confused.

             

            nVidia constantly updates their drivers, but despite their marketing efforts, it is up to Adobe to make use of these drivers by optimizing their MPE code, which is an on-going effort.

             

            AME export settings being slower than direct export is something I noticed with AVI exports, but I have not found any difference with H.264 exports and with MPEG2-DVD it is exactly the other way around. I have been discussing this with Adobe and they are aware of this issue with 5.02, but have not yet been able to replicate this issue with any reliability.

             

            I don't think Ann's answer is the solution. AFAIK there are no known issues with 'hacked' cards like the 480 versus the 470.

            • 3. Re: Odds and ends
              Jim_Simon Level 8
              have not yet been able to replicate this issue with any reliability.

               

              That's odd, because Tom's Hardware had no trouble duplicating it with every test they ran.  Here's a question they posed to Adobe:

               

              “If the direct export and render queue functions in Premiere Pro CS5 do the same job, but direct export is faster, why do we still have the queue in CS5?"  [Adobe] replied, “Because you can’t continue working when doing a direct export.”

               

              Notice there is no challenge to the issue.

              • 4. Re: Odds and ends
                Harm Millaard Level 7

                Here are some results that I experienced:

                 

                02-Sep-10 16-03-10.png

                 

                and that the Adobe engineer looking into this matter could not replicate, at least not consistently.

                 

                These figures are based on the PPBM5 tests with 5.02

                • 5. Re: Odds and ends
                  Jeff Bellune Adobe Community Professional
                  “If the direct export and render queue functions in Premiere Pro CS5 do the same job, but direct export is faster, why do we still have the queue in CS5?"

                  I have one word for the Tom's Hardware guys:

                   

                  Batch

                   

                  I'd like to see them set up a batch export without the queue.  If they can, then I will have learned something I didn't know before.

                   

                  -Jeff

                  • 6. Re: Odds and ends
                    Todd_Kopriva Level 8

                    > I have been discussing this with Adobe and they are aware of this issue with 5.02, but have not yet been able to replicate this issue with any reliability.

                     

                     

                    I had a pretty long conversation with one of the testers about this (difference in speeds between AME standalone encoding and direct export) yesterday, and he indicated that there appear to be several intertwined issues that mean that you get different results depending on various hardware variables. He said that he'd really appreciate people continuing to submit detailed bug reports on this issue, with details of hardware configurations, so that he can disentangle the various issues.

                     

                    One of the factors is NVIDIA drivers, and Adobe and NVIDIA are right now testing some new drivers that should address some of the issues underlying this discrepancy in speeds.

                    • 7. Re: Odds and ends
                      JaysonM-Y Level 3

                      LOL put an intern at the desk and leave em to export overnight.

                      • 8. Re: Odds and ends
                        Jim_Simon Level 8
                        I have one word for the Tom's Hardware guys:

                         

                        Batch

                         

                         

                        That was pretty much Adobe's response to the question.  My point was that no challenge to the issue of a speed difference was offered in the response, which indicates their awareness of it.

                         

                        The question is if Tom's can replicate it with every test they did, why can't Adobe?

                        • 9. Re: Odds and ends
                          Todd_Kopriva Level 8

                          > The question is if Tom's can replicate it with every test they did, why can't Adobe?

                           

                           

                          See my answer above.

                          • 10. Re: Odds and ends
                            Jim_Simon Level 8

                            I have to admit, it doesn't make much sense.  The functions do the same thing, using the same hardware and the same drivers.  Yet they perform differently.  That smacks of an internal code issue.

                            • 11. Re: Odds and ends
                              Mark Mapes Adobe Employee

                              I'm the QE who investigated these performances issues that Harm brought to our attention--and the unnamed individual whom both Harm and Todd referred to. Let's see if I can clarify regarding the ability (or inability) to reproduce. To stay as close as possible to comparing apples to apples, I used Harm's "PPBM CS5" project, sequences, and assets. Broadly speaking, that project consists of three sequences designed to test different aspects of performance. Following Harm's lead, I conducted 4 tests with each sequence, and I ran through the complete battery more than once on two computers with substantially different configurations, particularly with regards to number of cores and amount of RAM:

                              • Queue (AME)/HW GPU Enabled
                              • Queue (AME)/HW GPU Disabled
                              • Export Now/HW GPU Enabled
                              • Export Now/HW GPU Disabled

                               

                              I found considerable variation between Harm's results and mine, between my two test systems, and even to a lesser extent from one pass to the next on the same system. Here are just a few specifics. Note that Queue/GPU enabled is the baseline for the percentages provide, so a value lower than 100% means that the other scenario was faster (50% is twice as fast), while a value greater than 100% means the other scenario was slower (200% is twice as slow)

                              • "Disk Test" sequence, GPU enabled: Queue vs. Export Now
                                • Harm: 26%
                                • my T5500: 104%
                                • my 390: 55%
                              • "MPEG2-DVD" sequence, export via Queue: GPU enabled vs. disabled
                                • Harm: 63%
                                • my T5500: 113%
                                • my 390, first pass: 136%
                                • my 390, first pass: 157%

                              • "MPEG2-DVD" sequence, GPU enabled: Queue vs. Export Now
                                • Harm: 240%
                                • my T5500: 143%
                                • my 390, first pass: 120%
                                • my 390, first pass: 114%

                               

                              You can see that in some cases all the results leaned one direction but to substantially different degrees, while in others they were split.

                               

                              All of the results above were with PPRO 5.0.2/AME 5.0.1. More recently, I revisited this testing running the current code and with a new nVidia codec that's not yet released to the public, and here's the good news:

                              • "MPEG2-DVD" sequence, GPU enabled: Queue vs. Export Now: No Difference (and both times were notably faster than on 5.0.1/5.0.2)
                              • "MPEG2-DVD" sequence, GPU disabled: Queue vs. Export Now: No Difference ("Software only" was slower than with GPU enabled--175%)

                              Notice that I've made no attempt to explain the varying results--I'm most assuredly not the right person to venture into that terrain. Hopefully, though, this has at least shed some light as to what we've been able to replicate.

                              • 12. Re: Odds and ends
                                Harm Millaard Level 7

                                Mark,

                                 

                                Thank you for clarifying this.

                                 

                                I will be testing this again and mail you my results in the usual way.